OFFICE OF THE TEXAS STATE CHEMIST Texas Feed and Fertilizer Control Service Agriculture Analytical Service
Sources of variability in measuring aflatoxin and the role of sampling
Tim Herrman State Chemist and Director, Office of the Texas State Chemist
Texas A&M AgriLife Research Professor, Department of Soil and Crop Sciences
Texas A&M University
Presenter
Presentation Notes
Quantifying sources of variability involving aflatoxin measurement is key to risk management. My responsibility as state chemist and director of the office of the Texas State Chemist is to serve both consumers and the industry as the lead regulatory risk management for feed and fertilizer. The commercial feed control act and rules designate the authority of managing aflatoxin risk to the agency I direct. Texas has defined regulatory MLs for aflatoxin and fumonisin.
OFFICE OF THE TEXAS STATE CHEMIST
Sources of Variability
Sampling
Sample reduction
Sub-sample measurement
Testing platform
Analyst
Reagents
Detection
Sample matrix
Laboratory
Scale
Presenter
Presentation Notes
Sampling is frequently considered the greatest source of variability in aflatoxin measurement, illustrated in this diagram. Tom Whitaker and I have communicated back and forth on a proposed sampling standard for maize in the COMESA trade region of eastern and southern Africa. In his first email he said, “the sampling step accounts for a major (80%) portion of the total random error of the aflatoxin test procedure which leads to misclassification of lots (good lots rejected and bad lots accepted) by the sampling protocol.” I suspect most in the room are familiar with the Johansson Whitaker paper published in 2000 where they quantify the sources of variability from 16 bags of aflatoxin contaminated maize, each sample was 2.5 lbs resulting in 32 samples per bag (to represent sampling variability), each of which was subsampled twice and analyzed for aflatoxin (representing sample preparation variability), and each extract was analyzed twice (representing analytical variability), all by the same analyst using hplc. This work was contracted by the Grain Inspection Packers Stockyard Administration to quantify variability parameters for their current aflatoxin sampling scheme, which is 2.5 pounds per truck, 3 pounds per railcar, and 10 pounds per 5 car composite or ship hold. It is my understanding from visiting with Larry Freeze, the FGIS statistician, is was not GIPSAs intend to change their sampling protocol based on this work, rather, they were seeking statistical validation of their current practice. The development of performance curves used to identify the type I and type II error associated with sampling has subsequently been used to develop sampling standards for Codex and others.
OFFICE OF THE TEXAS STATE CHEMIST
Presenter
Presentation Notes
Sampling is a key component in controlling variability in aflatoxin measurement. In 2010, OTSC studied the sampling and aflatoxin testing performance of grain elevator personnel at 87 facilities during maize have and found that 18 percent of the commercial grain elevators collected 5 or more probes from a truck delivering maize during harvest.
OFFICE OF THE TEXAS STATE CHEMIST
Presenter
Presentation Notes
The levels of aflatoxin found in Texas can present significant challenges to managing food safety risk. We map the level of aflatoxin by county, designed by color code based on the highest level measured by our agency. OTSC found in 2010 that one of more trucks contained greater that 1000 ppb in 7 Texas counties in 2010, 8 counties with greater than 300 to 500 ppb, 24 counties with greater than 20 to 300, and 26 counties where we did not encounter trucks containing over 20 ppb aflatoxin. During 2010, maize crops along the I35 corridor received a timely rain resulting in many field experiencing yields ranging from 120 to 140 percent of normal. This created a tremendous tension in the state, where farmers would receive crop insurance payments around 50% of the potential revenue verses selling the maize above 300 ppb, where they could receive 100% indemnification for the amount insured.
OFFICE OF THE TEXAS STATE CHEMIST
Approved Equipment & Procedures
Training for Individual Employees
Proficiency Verification
Process
Management & Recordkeeping
Monitoring & Corrective
Actions
Standardized methods Standardized training Verification of
employee performance Documented program
outcomes Monitoring &
corrective actions Reduced market and
food safety risk
One-Sample Strategy Program Components
Presenter
Presentation Notes
OTSC was working on and completed a plan in 2010 to work with the Risk Management Agency and grain elevators captured in a WHITE PAPER ON AFLATOXIN RISK MANAGEMENT IN TEXAS PURSUIT OF A ONE SAMPLE STRATEGY that was shared with legislators and industry. This strategy depicted on this slide resembles a plan-do-check-act Shewart cycle. I will discuss each of the steps in the program involving approved equipment and procedures, training for individual, proficiency testing and verification, recordkeeping and our agency’s monitoring. The five areas results in standardized methods and training and overall reduced market and food safety risk. The title One-Sample conveys the concept that one sample correctly collected, ground and tested using approve methods will yield an acceptable result. As a result of grain elevators in Texas incorporating this program, one sample is used for purchasing, regulatory and crop insurance purposes. We refer to this as co-regulation which I would offered up as a potential topic for the next world mycotoxin forum.
PATTERN 1: 7 probes for trucks or trailers loaded with grain more than 4 feet deep
PATTERN 2: 9 probes for trucks or trailers loaded with grain less than 4 feet deep
Presenter
Presentation Notes
Sampling and sample preparation protocol following GIPSA procedures are followed with several minor exceptions including the use of a 5 pound sample
OFFICE OF THE TEXAS STATE CHEMIST
Criteria: Grinding
Grind the entire sample
Collect at least 500 grams of the ground sample
70% of the particles pass through a 20 mesh sieve after grinding
Presenter
Presentation Notes
and monitoring fineness of grind using a 20 mesh sieve.
OFFICE OF THE TEXAS STATE CHEMIST
80 ppb Acceptable range:
60 - 100 ppb
Company A
Company B
Control Chart
Presenter
Presentation Notes
Results for the reference material that we provide to program participants is evaluated, as demonstrated by these control charts and one-sample strategy firms use validated test kit approved for used in the program. During the last global mycotoxin forum, Dr. Dai from the Office of the Texas State Chemist explained our agency’s program to validate and approve aflatoxin test kits. I will not discuss test kit validation in this presentation except for the use of results used to estimate intermediate precision with an equation used to calculate uncertainty.
Retained sample analysis in an ISO 17025 accredited lab
Presenter
Presentation Notes
Employee qualification and proficiency is assessed each year, equipment performance including grinder and scale are monitored by the firm, recorded and verified by our field investigators who also collect retained samples that we verify in our ISO 17025 accredited lab in College Station Texas.
OFFICE OF THE TEXAS STATE CHEMIST
Performance curve for 2013-2015
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 20 40 60 80 100 120 140
Acc
epta
nce
Pro
bab
ility
Aflatoxin Conc. (ppb)
Presenter
Presentation Notes
OTSC verified approximately 1% of the one-sample strategy participants samples, these results are graphed using a performance curve procedure. Using the regression curve, 4% of the population is type I error (above the curve and to the left of the 20 ppb aflatoxin concentration, 16% of the population is type II error (below the curve and to the right of the 20 ppb aflatoxin concentration, and 80% is correctly assigned as either above or below the Texas regulatory ML when comparing OTSC results on the x-axis against one-sample strategy firms acceptance probability on the y (vertical) axis.
OFFICE OF THE TEXAS STATE CHEMIST
SAMPLING
Inference about the population
Presenter
Presentation Notes
Sampling enables use to make an inference about the population without having to measure everything.
OFFICE OF THE TEXAS STATE CHEMIST Variance Structure of Aflatoxin Contaminated Maize in
Commercial Grain Elevators and Transporters Variance
Source Percent of
Total Variance
Facility 1.9
Bin 65.8
Truck 9.1
Sampling and Testing Error
23.2
Herrman et al. JRS 1(1):23-31
Presenter
Presentation Notes
OTSC evaluated the variance structure of aflatoxin concentration from multiple facilities, bins, trucks, and probes (captured in the error term), all of which could be characterized as different sources of sampling variability. We did not consider subsamples or rerun extracts. All facilities were more or less uniformly contaminated, but bins with in facility contributed the greatest source of variability. Perhaps of interest is the error term which is a composite of variability attributed to individual probes and analytical variability. Seven single probe (250 g) samples were collected from each truck. From this variance component analysis, sample variability within a truck explained about 23% of the total aflatoxin variability among 7 grain elevators, 38 storage bins and 64 trucks.
OFFICE OF THE TEXAS STATE CHEMIST Variance Structure of Aflatoxin Contaminated Maize in
Commercial Maize Mills in Kenya Variance
Source Percent of
Total Variance
Mill 0
Truck 7.8
Bag 33.3
Within bag 50.1
Analytical 3.4
Error 5.4
Presenter
Presentation Notes
OTSC examined the variance structure of aflatoxin contaminated maize delivered to three commercial mills in Kenya and summarized the variability between bags from different trucks and mills at our ISO 17025 lab located on the ILRI campus in Nairobi. Variability within bags and between bags closely aligns with the Johansson Witaker paper and followed a similar experimental design, only these results were generated from commercial movement of maize into the formal milling sector. Variability also seems greater when looking at individual bags from different sources (e.g. farms in Kenya) than bulk grain moving in the market, typical of the US. In summary, these results indicate that: when designing a study to examine variability attributed to sampling, we find sampling is the major source of variability. When looking at the variability or an entire grain harvest or commercial system, it accounted for a little over 20% of the variability.
OFFICE OF THE TEXAS STATE CHEMIST
GRINDING
Retaining the representative property of the sample
Presenter
Presentation Notes
Grinding procedures determine whether the representative properties of the sample are retained or lost.
OFFICE OF THE TEXAS STATE CHEMIST
Sample Grinding
Presenter
Presentation Notes
The entire sample needs to be ground and reduced. OTSC double grinds samples using a Romer RAS mill
OFFICE OF THE TEXAS STATE CHEMIST
Sample Grinding
Presenter
Presentation Notes
The second grind is performed using a Retch
OFFICE OF THE TEXAS STATE CHEMIST
Presenter
Presentation Notes
With a diameter opening of the screen 1.5 mm. This procedure yields over 95 of the material through a 20 mesh sieve, on the criteria established by GIPSA for proficiency testing protocol. As illustrated in this slide, cleaning the sieves and grinder between samples is essential.
OFFICE OF THE TEXAS STATE CHEMIST
Presenter
Presentation Notes
Further mixing of the ground sample is performed in our lab using 4 corner mixing prior to placing the sample in a container for laboratory analysis.
OFFICE OF THE TEXAS STATE CHEMIST
REFERENCE MATERIAL
Developing uniform working controls
Presenter
Presentation Notes
OFFICE OF THE TEXAS STATE CHEMIST
Creating Reference Material
Presenter
Presentation Notes
We produce reference material for our lab and participants in the One Sample Strategy program in Texas as well as reference material for our APTECA program that is currently being used in 4 east African countries, as well as our proficiency testing program reaching labs on every continent but Antarctica.
OFFICE OF THE TEXAS STATE CHEMIST
Recommendation 9: Sufficient Homogeneity In testing for sufficient homogeneity, duplicate results from a single distribution unit should be deleted before the analysis of variance if they are shown to be significantly different from each other by Cochran’s test at the 99% level of confidence
Presenter
Presentation Notes
We use the international harmonize protocol for the proficiency testing of analytical chemistry laboratories procedure to assess homogeneity.
OFFICE OF THE TEXAS STATE CHEMIST
Sufficient stability Changes in test material are inconsequential Period in question is the interval between preparation of the material and the deadline for return of the results 5 samples will be analyzed after the proficiency test
Presenter
Presentation Notes
We also have evaluated some of our reference material placed in Kenya as well as our proficiency testing material for sufficient stability to prepare our program for ISO 17043 accreditation, which we hope to achieve later this year.
OFFICE OF THE TEXAS STATE CHEMIST
UNCERTAINTY & VARIABILITY
Laboratory uncertainty
Presenter
Presentation Notes
Laboratory variability is known as uncertainty within the laboratory community.
OFFICE OF THE TEXAS STATE CHEMIST
Uncertainty
ISO 17025 5.4.6.2 Testing laboratories shall
have and shall apply procedures for estimating uncertainty of measurement…
Reasonable estimation shall be based on knowledge of the performance of the method and on the measurement scope and shall make use of, for example, previous experience and validation data
Uncertainty Budget List all potential factors
affecting variability in measurements –make table
Determine the standard uncertainty for each factor including distribution
Perform root sum squares for all factors to create the combined or standard uncertainty
Multiply by coverage factor: 2
𝑆𝐼 = 𝑆𝑎2 + 𝑆𝑏2 … 𝑆𝑥2
Presenter
Presentation Notes
Several methods are available to calculate uncertainty including a budget technique and, perhaps more common in the US, calculating the relative standard deviation of the labs working control (or reference material) run daily to ensure accurate results. We perform both, the uncertainty budget is required by the Kenya Accreditation Service (KENAS) for our ISO 17025 accredited lab in Kenya and we use the RSD method published by A2LA in the States for our ISO 17025 accredited lab in College Station TX.
This table illustrates the uncertainty calculation for aflatoxin and fumonisin in our lab with a coverage factor of 2 based on our working controls.
OFFICE OF THE TEXAS STATE CHEMIST
PROFICIENCY TEST RESULTS
One of the Big Three
Presenter
Presentation Notes
OTSC delivers proficiency testing to the global community in collaboration with FAO and others through our program titled: Aflatoxin proficiency testing and control in Africa, Asia and the Americas, which I refer to as APTECA. Proficiency testing helps evaluate the performance of laboratories, indicates problems in the lab which may be related to inadequate test or measurement procedures, effectiveness of staff training or calibration of equipment.
OFFICE OF THE TEXAS STATE CHEMIST
Presenter
Presentation Notes
The samples are ground maize and are prepared using protocol I described earlier.
OFFICE OF THE TEXAS STATE CHEMIST
Proficiency Testing Performance
0%10%20%30%40%50%60%70%80%90%
1 2 3 4 5
RSD
Proficiency Sample Number
Presenter
Presentation Notes
These are results from one subset of the laboratories performing aflatoxin analysis in Africa. Proficiency test 1 is the first proficiency test yield a RSD of 78%, the second proficiency test didn’t see any improvement. Proficiency sample 3 is the same material used in round 2, however, it was tested using the slurry method from a 250 g sample and yielded a 68% RSD. After training and analyst qualification in Nairobi, the same people produced RSDs of to 32 and 16 percent for two testing platforms approved for use in the APTECA program from a common extract. Thus: when designing a study to examine variability attributed to lab analysis, we find lab analysis is the major source of variability.
OFFICE OF THE TEXAS STATE CHEMIST
LABORATORY ENVIRONMENT
APTECA group qualification exercises
Presenter
Presentation Notes
The environment of a lab is more critical as instrumentation become more technical, such as the use of HPLC or UPLC/MS technology.
OFFICE OF THE TEXAS STATE CHEMIST
OFFICE OF THE TEXAS STATE CHEMIST
SUMMARY
Sources of Variability
Presenter
Presentation Notes
In this presentation, I have discussed the source of variability that we have measured.
OFFICE OF THE TEXAS STATE CHEMIST
Material Method Environment
Operator Machine
Aflatoxin Measurement
Cause and Effect Diagram
Matrix Reagent
Reference Material
Sampling Grinding
Extraction Validation
Analyst
Reader Scale
Grinder
Presenter
Presentation Notes
This diagram includes other sources of variability that we have observed but not measured using a hierarchical design similar to those in our variance component analysis. In Africa, for example, obtaining pure water is a challenge as well as grinding, since there is a flint maize and much harder that found in the US. My focus has been on maize, however, as an agency that regulate animal feed we are will aware of the matrix effect found in some ingredients and products.
OFFICE OF THE TEXAS STATE CHEMIST
Uncertainty Budget for Total Variability
Sampling CV = 82%
Test method CV = 46%
Analyst CV = 32%
Uncertainty Budget
𝑆𝐼 = 82𝑎2 + 46𝑏2 + 32𝑐2
𝑆𝐼 = 𝑆𝑎2 + 𝑆𝑏2 … 𝑆𝑥2
Sources of Variability
𝑆 = 99%
Presenter
Presentation Notes
Using an uncertainty budget for total variability, I have considered worse and best case scenarios. For example, the Johansson Witaker paper reports a CV of 82% for variability and our proficiency study resulted in a 78% CV, which I have further partitioned into analysts at 32 percent and taking a forced weight of 46% for test method, which actually coincided with the RSD values we have obtained for some aflatoxin test kits in our validation study. Using the root square method described earlier, the uncertainty is 99%. With a coverage factor of 2, this would indicate that our uncertainty for any measurement (or confidence interval) is around 200% at a 95% level of significance.
OFFICE OF THE TEXAS STATE CHEMIST
Uncertainty Budget for Total Variability
Sampling CV = 23%
Test method CV = 16%
Analyst CV = 16%
Uncertainty Budget
𝑆𝐼 = 23𝑎2 + 16𝑏2 + 16𝑐2
𝑆𝐼 = 𝑆𝑎2 + 𝑆𝑏2 … 𝑆𝑥2
Sources of Variability
𝑆 = 32%
Presenter
Presentation Notes
In the best case scenario, I used the sampling variability from the grain elevator study, 16% testing variability which is within the GIPSA protocol as acceptable, and an analyst CV or 16%, a proficiency test result after the group qualification exercise and what we commonly observe among participants in either the one-sample or APTECA program. The total uncertainty is 32% using the root square calculation and with a coverage factor of 2 yield a 64% level of uncertainty in measuring aflatoxin in a commercial load of maize.
OFFICE OF THE TEXAS STATE CHEMIST
SOURCES OF VARIABILITY IN MEASURING AFLATOXIN AND THE ROLE OF SAMPLING A continuous improvement approach to define, measure, and control aflatoxin helped reduce food safety risk.
Presenter
Presentation Notes
In summary, implementing a continuous improvement approach to define, measure, and control aflatoxin has helped reduce food safety risk in Texas and Kenya as well as our new collaborators in Uganda, Rwanda, and Tanzania.
OFFICE OF THE TEXAS STATE CHEMIST Texas Feed and Fertilizer Control Service Agriculture Analytical Service
Acknowledgements Office of the Texas State Chemist Personnel
Cindy McCormick, Carlton Peterson K.M. Lee, Susie Dai, Anne Muiruri
Quantifying sources of variability involving aflatoxin measurement is key to risk management. As a advocate of continuous improvement, I frequently cite a saying I learned years ago, “you can’t improve what you do not control, you can’t control what you do not measure, and you can’t measure what you do not define. My responsibility as state chemist and director of the office of the Texas State Chemist is to serve both consumers and the industry as the lead regulatory risk management for feed and fertilizer. The commercial feed control act and rules designate the authority of managing aflatoxin risk to the agency I direct. Texas has defined regulatory MLs for aflatoxin and fumonisin.